CN106772032A - A kind of fault signature extracting method of turbine-generator units - Google Patents

Abstract

The invention discloses a kind of fault signature extracting method of turbine-generator units, the method is based on two-dimensional empirical mode decomposition and vector spectrum is analyzed, consider the advantage and disadvantage of empirical mode decomposition method and the vibration signal situation of turbine-generator units, solve the problems, such as that conventional method is difficult to obtain comprehensively and accurately extraction result.The inventive method can detect the characteristic frequency of the early signal that is out of order from the complex vibration signal of turbine-generator units actual measurement comprehensively, exactly.And the reliability of diagnostic result is higher, operation maintenance personnel is facilitated to process this failure in time, it is to avoid the generation of accident, so as to ensure the safety economy Effec-tive Function of whole system.

Description

A kind of fault signature extracting method of turbine-generator units

Technical field

The invention belongs to fault diagnosis technology field, and in particular to a kind of fault signature extraction side of turbine-generator units Method.

Background technology

With the development of China's waterpower industry, water turbine set gradually develops towards the direction that single-machine capacity is big, operating mode is complicated, The generation of failure and development include substantial amounts of uncertain factor in its running, and the vibration signal of the hydraulic turbine is often showed For it is non-linear, non-stationary the characteristics of.The vibration signal of water turbine set can reflect the running status of turbine-generator units, therefore The vibration signal of analysis and research water turbine set is significant.

But complexity, diversity, coupling property and uncertainty due to Hydropower Unit failure, a kind of failure there may be Multi-party region feature and sign, influence each other and restrict between various faults factor, cause and contain in unit fault vibration signal The fault characteristic information of mutual aliasing, thus using single channel carry out signal analysis and feature extraction often do not reach it is satisfied Effect, this is the problem that must be solved.

In the failure diagnostic process for carrying out the hydraulic turbine, typically using Time-Frequency Analysis Method, such as Short Time Fourier Transform, The methods such as Wigner-Ville distribution, wavelet transformation and empirical mode decomposition.FFT, Wigner-Ville distribution method are all ratios Relatively it is applied to linear signal, just seems extremely difficult for processing this non-stationary signal, unsuitable treatment water turbine set This nonlinear properties.In recent years, after the proposition of wavelet analysis, just widely accepted and applied, but it also brings along Some problems, such as Selection of Wavelet Basis are difficult, parameter sensitivity and stationarity hypothesis etc., and the result to nonlinear properties is not ten Sub-argument is thought.Empirical mode decomposition is a kind of new Time-Frequency Analysis Method, because it can be carried out according to the yardstick of setting difference The decomposition of self adaptation, thus with very strong adaptivity, be particularly suitable for dividing this non-stationary signal of water turbine set Analysis and feature extraction.But empirical mode decomposition can only process one-dimensional signal, and there are problems that end effect, Thus extraction effect is often not ideal enough.

The content of the invention

It is an object of the invention to provide a kind of fault signature extracting method of turbine-generator units, the method is based on two dimension and passes through Mode decomposition and vector spectrum analysis are tested, the vibration of the advantage and disadvantage and turbine-generator units of empirical mode decomposition method is considered RST, solves the problems, such as that conventional method is difficult to obtain comprehensively and accurately extraction result.

The technical solution adopted in the present invention is, a kind of fault signature extracting method of turbine-generator units, including following Step：

Step 1：Horizontal and vertical primary signal x (t), y are obtained using the vibrating sensor of water-turbine unit (t), so as to obtain complex signal z (t)=x (t)+iy (t)；

Step 2：Determine projecting direction

Step 3：Complex signal z (t) is projected toOn, obtain

Step 4：ExtractLocal maximum when corresponding momentThen to setCarry out Interpolation, obtains in directionOn maximum envelope；

Step 5：Calculate barycenter m (t) corresponding to maximum envelope in all directions；

Step 6：S (t)=z (t)-m (t) is calculated, and judges whether S (t) meets the condition of IMF, if it is satisfied, then making S_i T ()=S (t), is transferred to step 7；If it is not satisfied, then make z (t)=S (t), then repeat step 3-6, until meeting condition；

Step 7：I-th IMF component is isolated from signal；

m_i(t)=z (t)-S_i(t)

Judge m_iWhether t () be monotonic function, if it is, circulation terminates, obtains the n IMF component for meeting condition；Such as Fruit is not then to make z (t)=m_iT (), goes to step 3；

Step 8：Each rank natural mode of vibration component IMF that will be obtained_i(i=1,2 ..., n) it is divided into real part IMF1_i(i=1, 2 ..., n) with imaginary part IMF2_i(i=1,2 ..., n), calculate the mutual information of the corresponding primary signal of each modal components；

Step 9：Normalized is done to mutual information；

Step 10：Screening modal components.Selected threshold, using the mutual information of modal components and original signal less than threshold value as Chaff component is rejected, and modal components are reconstructed with the mutual information of original signal more than the component of threshold value；

Step 11：The sequence that obtains will be reconstructed and constitute one group of complex sequences, and complex sequences to constructing carries out Fourier changes Change；

Step 12：The vector spectrum of analytical sequence is calculated, the fault signature of turbine-generator units is obtained by vector spectrum figure.

The features of the present invention is also resided in：

Step 8 is specially：Each rank natural mode of vibration component IMF that will be obtained_i(i=1,2 ..., n) it is divided into real part IMF1_i(i= 1,2 ..., n) with imaginary part IMF2_i(i=1,2 ..., n), calculate component IMF1_iAnd IMF2_iPrimary signal x (t), the edge of y (t) Probability distribution p (IMF1_i)、p(IMF2_i), p (x), p (y), calculate real component IMF1_iWith the joint probability of primary signal x (t) Distribution is respectively p (IMF1_i, x), imaginary IMF1_iWith the joint probability p (IMF2 of primary signal y (t)_i, y), so as to obtain The mutual information of the corresponding primary signal of each modal components

The beneficial effects of the invention are as follows, the fault signature extracting method of turbine-generator units proposed by the present invention, can comprehensively, The characteristic frequency of the early signal that is out of order is detected from the complex vibration signal of turbine-generator units actual measurement exactly.And diagnosis knot The reliability of fruit is higher, facilitates operation maintenance personnel to process this failure in time, it is to avoid the generation of accident, so as to ensure whole The safety economy Effec-tive Function of individual system.

Brief description of the drawings

Fig. 1 is actual measurement runner of hydro-generating unit vibration signal waveforms figure in the present invention；

Fig. 2 is the spectrogram of vibration signals measured X-direction signal of the present invention；

Fig. 3 is the spectrogram of vibration signals measured Y-direction signal of the present invention；

Fig. 4 is the method flow diagram that fault signature of the present invention is extracted；

Fig. 5 is the decomposition result of vibration signals measured BEMD of the present invention；

Fig. 6 is the decomposition result of vibration signals measured X-direction BEMD of the present invention；

Fig. 7 is the decomposition result of vibration signals measured Y-direction BEMD of the present invention；

Fig. 8 is the vector spectrum that the present invention screens the reconstruction signal for obtaining by mutual information.

Specific embodiment

The present invention is described in further detail with reference to the accompanying drawings and detailed description, but the present invention is not limited to These implementation methods.

Below so that the fault signature of certain turbine-generator units is extracted as an example.In the turbine-generator units bulb structure (peace Mounted in other structures also can) while installing horizontal and vertical two vibrating sensors, two sampling lengths of sensor are 1024 Individual, sample frequency is 229HZ, and the rated speed of water turbine set is 107.1r/min, and maximum head is 25.7m, rated head 16.1m, the power of the hydraulic turbine is 49MW.

Horizontal and vertical primary signal x (t), y are obtained first with the vibrating sensor of water-turbine unit T (), the primary signal in both direction is as shown in Figure 1.Then the primary signal to both direction carries out Fourier transformation, obtains Fig. 2 and Fig. 3.In Fig. 2, it is characteristic frequency that unit has in itself to turn frequency and 100Hz frequencies, in addition, also with energy 10 frequencys multiplication higher and the signal of 15 frequencys multiplication.Fig. 3 transfers frequency has energy higher with the frequency of 100HZ, and this reflects with Fig. 2 Result is consistent, and has energy 15 frequencys multiplication higher simultaneously, and this is also consistent with the result of Fig. 2.It is different from Fig. 2 It is that there is energy low frequency signal higher in Fig. 3, and without 10 frequency-doubled signals.It can be seen that the signal that both direction is extracted is special It is inconsistent to levy, and the result of signal characteristic abstraction often influences the judgement to unit malfunction.

Be can be seen that by the above results, the result that different channel characteristics are extracted is inconsistent.Therefore fault diagnosis of the present invention is used Method is diagnosed to turbine-generator units, as shown in figure 4, in turbine-generator units normal course of operation, by following step Suddenly fault diagnosis is carried out.

Step 1：Horizontal and vertical primary signal x (t), y are obtained using the vibrating sensor of water-turbine unit (t), so as to obtain complex signal z (t)=x (t)+iy (t).

Step 2：Determine projecting directionWherein 1≤k≤M, M are the projecting direction number of setting, and the value is got over Greatly, effect is more preferable, and M=32 is set herein.

Step 3：Complex signal z (t) is projected toOn, obtain

Step 4：ExtractLocal maximum when corresponding momentThen to setInserted Value, obtains in directionOn maximum envelope

Step 5：Calculate barycenter m (t) corresponding to maximum envelope in all directions

Step 6：Calculate

S (t)=z (t)-m (t) (3)

And judge whether S (t) meets the condition of IMF, if it is satisfied, then making S_iT ()=S (t), is transferred to step 7.If discontented Foot, then make z (t)=S (t), then repeat step 3-6, until meeting condition.Here the method for judging whether to meet IMF conditions It is the standard similar to Cauchy convergence criterions be given by Huang, it defines following standard deviation：

Generally by SD values be 0.2 to 0.3 between, that is, meet 0.2<SD<Screening process just terminates when 0.3.This standard Physical significance is：S should be caused_iT the requirement of () close enough IMF, controls the number of times of screening so that resulting again IMF components retain the information of amplitude modulation in primary signal.

Step 7：I-th IMF component is isolated from signal；

m_i(t)=z (t)-S_i(t) (5)

Judge m_iWhether t () be monotonic function, if it is, circulation terminates；If it is not, then making z (t)=m_i(t)

Go to step 3.

At the end of circulation, the n IMF component for meeting condition, this seasonal r have been obtained_n=m_n(t), referred to as residual components. Decomposition result can so be obtained is：

Residual components r in formula_nRepresent the average tendency of signal；And each IMF components S₁(t),S₂(t),…,S_n(t) difference The composition of signal different frequency sections from high to low is contained, the frequency content that each frequency band is included is different, and Change with signal conversion in itself.

Decomposed using the signal collected from hydraulic turbine bulb vibrating body mentioned above, the result for obtaining such as Fig. 5 Shown, wherein solid line represents real component, and imaginary part represents imaginary.The method can simultaneously be carried out the signal of both direction The phase information and synchronism of empirical mode decomposition, X-direction and Y-direction signal can be effectively maintained.And what is decomposed During take into account correlation between X-direction signal and Y-direction signal, thus decomposition result can preferably retain letter Number feature, and with good synchronism.For the ease of observing the discomposing effect of the method, by X and the decomposition result of Y-direction It is individually placed in different figures, decomposition result is shown in Fig. 6 and Fig. 7.It is contemplated that the result of empirical mode decomposition often has void False component, the present invention by calculating the mutual information between each component and original signal, by given threshold, filter out mutual information compared with Big component.And the component to filtering out carries out the reconstruct of signal, one group of new complex sequences is obtained.

Step 8：Each rank natural mode of vibration component IMF1 that will be obtained_i(i=1,2 ..., n) it is divided into real part IMF1_i(i=1, 2 ..., n) with imaginary part IMF2_i(i=1,2 ..., n).Calculate component IMF1_iAnd IMF2_iPrimary signal x (t), the edge of y (t) is general Rate distribution p (IMF1_i)、p(IMF2_i)、p(x)、p(y).Calculate real component IMF1_iWith the joint probability point of primary signal x (t) Cloth is respectively p (IMF1_i, x), imaginary IMF1_iWith the joint probability p (IMF2 of primary signal y (t)_i,y).It is each so as to obtain The mutual information of the corresponding primary signal of individual modal components

Step 9：Normalized is done to mutual information.

β=MI_i/max(MI_i) (9)

Step 10：Screening modal components.The threshold value for choosing β is 0.02, and modal components are less than with the mutual information of original signal Threshold value is rejected as chaff component.And the mutual information of modal components and original signal is carried out into weight more than the component of threshold value Structure.Obtain reconstruction signalWith

IMF1 ' in formula_iRepresent the component of the mutual information more than threshold value of real part modal components and original signal, IMF2 '_iRepresent void Portion's modal components are more than the component of threshold value with the mutual information of original signal.

Step 11：The sequence for obtaining will be reconstructedWithConstitute one group of complex sequencesAnd the complex sequences to constructing enters Row Fourier is converted, and obtains frequency domain complex signal

So as to obtain

In formulaRespectively sequence X_ciAmplitude and phase angle；Respectively sequence Y_ciAmplitude and phase angle.I=1,2 ..., N/2-1, its Middle N represents sampling length.

Step 12：Calculate the vector spectrum of analytical sequence.If R_aiIt is oval major axis or maximum intensity, shake arrow based on definition, R_biIt is the vertical direction of oval short axle or maximum intensity, is defined as pair and shakes arrow, a_iBased on shake the angle sweared in X-axis, φ_iFor this The Initial phase of elliptical orbit under frequency.Drawn by formula (13)

The characteristic information of signal can preferably be reflected by vector spectrum figure.Vector spectrum figure such as Fig. 8 of examples detailed above data It is shown.The characteristics of in view of water turbine set signal, retain the master in vector spectrum analysis and shake arrow figure and phase place change figure.The main arrow figure that shakes In addition to it can reflect signal turn frequency in itself and 100Hz frequencies, both direction is also extracted into different characteristic frequencies and is reflected Come, additionally, original, in different spectral figure energy, 5 prominent frequency-doubled signals are not reflected well herein.By the hydraulic turbine Failure mechanism understand, unit in actual motion, when unit run with Smaller load, vibrations aggravation, pressure fluctuation increasing feelings Under condition, produce vortex rope, vortex rope phenomenon to cause bearing wear near draft tube, 5 a small amount of frequency-doubled signals will be produced, due to The signal is fainter in itself, and suffers from the influence of strong background noise, and signal characteristic is difficult to be extracted, and of the invention The signal characteristic can be reflected well.Phase place change figure may be used to determine the initial phase of failure.As can be seen here, The method can efficiently extract the feature of water turbine set signal.The method not only greatly reduces calculating in terms of numerical computations Amount, while being become apparent from spectrum analysis accurately, so that engineers and technicians are more easy to judge fault message.

As can be seen here, compared to traditional method, set forth herein two-dimensional empirical mode decomposition and vector spectrum analysis spy Levy extracting method, can more fully, the vibration signal that is accurately detected in hydrogenerator, and diagnostic result is more true It is reliable, the feature of water turbine set signal can be efficiently extracted, for the fault signature of electrical equipment provide it is a kind of newly Thinking.

Claims (2)

1. the fault signature extracting method of a kind of turbine-generator units, it is characterised in that comprise the following steps：

Step 1：Horizontal and vertical primary signal x (t) is obtained using the vibrating sensor of water-turbine unit, y (t), So as to obtain complex signal z (t)=x (t)+iy (t)；

Step 2：Determine projecting direction

Step 3：Complex signal z (t) is projected toOn, obtain

Step 4：ExtractLocal maximum when corresponding momentThen to setEnter row interpolation, Obtain in directionOn maximum envelope；

Step 5：Calculate barycenter m (t) corresponding to maximum envelope in all directions；

Step 6：S (t)=z (t)-m (t) is calculated, and judges whether S (t) meets the condition of IMF, if it is satisfied, then making S_i(t)= S (t), is transferred to step 7；If it is not satisfied, then make z (t)=S (t), then repeat step 3-6, until meeting condition；

Step 7：I-th IMF component is isolated from signal；

m_i(t)=x (t)-S_i(t)

Judge m_iWhether t () be monotonic function, if it is, circulation terminates, obtains the n IMF component for meeting condition；If no It is then to make x (t)=m_iT (), goes to step 3；

Step 8：Each rank natural mode of vibration component IMF that will be obtained_i(i=1,2 ..., n) it is divided into real part IMF1_i(i=1,2 ..., n) With imaginary part IMF2_i(i=1,2 ..., n), calculate the mutual information of the corresponding primary signal of each modal components；

Step 9：Normalized is done to mutual information；

Step 10：Screening modal components, selected threshold, using the mutual information of modal components and original signal less than threshold value as falseness Component is rejected, and modal components are reconstructed with the mutual information of original signal more than the component of threshold value；

Step 11：The sequence that obtains will be reconstructed and constitute one group of complex sequences, and complex sequences to constructing carries out Fourier conversion；

Step 12：The vector spectrum of analytical sequence is calculated, the fault signature of turbine-generator units is obtained by vector spectrum figure.

2. the fault signature extracting method of turbine-generator units according to claim 1, it is characterised in that the step 8 Specially：

Each rank natural mode of vibration component IMF that will be obtained_i(i=1,2 ..., n) it is divided into real part IMF1_i(i=1,2 ..., n) and imaginary part IMF2_i(i=1,2 ..., n), calculate component IMF1_iAnd IMF2_iPrimary signal x (t), the marginal probability distribution p of y (t) (IMF1_i)、p(IMF2_i), p (x), p (y), calculate real component IMF1_iJoint probability distribution with primary signal x (t) is respectively p(IMF1_i, x), imaginary IMF1_iWith the joint probability p (IMF2 of primary signal y (t)_i, y), so as to obtain each mode point The mutual information of the corresponding primary signal of amount

$MI\left(IMF{1}_i,x\right)=\underset{IMF{1}_i}{\mathrm{\&Sigma;}}\underset{x}{\mathrm{\&Sigma;}}p\left(IMF{1}_i,x\right)\log \frac{p\left(IMF{1}_i,x\right)}{p\left(IMF{1}_i\right)p\left(x\right)}$

$MI(IMF{2}_i,y)=\underset{IMF{2}_i}{\&Sigma;}\underset{y}{\&Sigma;}p(IMF{2}_i,y)log\frac{p(IMF{2}_i,y)}{p\left(IMF{2}_i\right)p\left(y\right)}.$